Journal of Science and Development April 2008: 31-43 HANOI UNIVERSITY OF AGRICULTURE Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam Nguyen Thi Thu Ha * *Center for Agricultural Research and Ecological Studies (CARES), Hanoi University of Agriculture Abstract The need for land use and land cover change information has become a focus in current strategies for managing natural resources and monitoring environmental change. In order to investigate the underlying causes of forest cover change over the period 1998 - 2003 in the two upper-most districts of the Ca River Basin, remote sensing data was used together with the multiple logistic regression technique. Supervised classification of Landsat imagery captured in 1998 and 2003 was performed and the findings show that over five years the total area of forest cover change was about 12,400 ha, of which the total area of forest regrowth was 76,000 ha. The subsequent analysis of the driving forces behind these changes by using the multiple logistic regression technique proved that the Forest Land Allocation policy and natural management practices by humans were the most important factors. These factors were reflected through the number of livestock per area, population density, and elevation in the forest regrowth model; in the model of deforestation they were the implementation process of the land allocation policy, food security, and livestock density. These predictors have created a very good logistic model for forest cover changes with the ranging from 0.22 to 0.68. 2 L R Keywords: Driving forces, Land cover, Ca River Basin, Vietnam 1. INTRODUCTION Tropical forests are nature’s most extravagant gardens. Straddling the equator in three major regions: Southeast Asia, West Africa, and South and Central America, tropical rain forests are home to many rain forest species and account for approximately 50% of the world’s biodiversity (Goldsmith, 1998; Molles, 2002). The global distribution of tropical rain forests corresponds to areas where conditions are warm and wet year-round with the average temperature around 25 0 C to 27 0 C and an annual rainfall range of 2,000 to 4,000 mm. These conditions are ideal for creating one of the richest ecosystems on Earth. The rapid destruction of tropical rain forest has been recognized as a major contributor to global warming (Fearnside, 2000; Nascimento & Laurance, 2002). Tropical rainforest destruction is the result of agricultural land expansion, urbanization, logging, and other types of human intervention. In Vietnam, a dramatic change in the amount of forest cover was experienced during the second half of the 20 th century (Do Dinh Sam et al.). During this period, forests were reduced from comprising 33.8% of the country’s land mass (about 330,000 km 2 in total) in 1976 to 30.1% in 1985 and to 28.2% in 1995 (Do Dinh Sam et al.). The Ca River Basin is located in Nghe An province in central Vietnam. The basin covers a vast area of about 828,357 ha and spreads over 8 provincial districts, of which 5 districts (Ky Son, Tuong Duong, Con Cuong, Anh Son and Thanh Chuong) are considered the upland part of the basin. This region has 31 Nguyen Thi Thu Ha long been considered as having the richest area of forest cover in the country. To protect the forest area in this region, the government launched and implemented a number of national programs (e.g. PAM, Program 327, and an ongoing 5 million hectare reforestation program) (Nguyen Thi Thu Ha, 2001). These programs are aimed at providing the local communities environmentally sound production alternatives, and thus reducing the pressure on local forests. However, as the area’s population grows, increased demand for land for agricultural cultivation has put more pressure on the forest. Local communities are mostly poor and dependent on forest resources for supplementary sources of income, especially in the event of crop failures and during the transition time between the two annual harvests. Forests are also the dominant sources of household energy for cooking, construction materials, animal fodder and traditional medicines (Nguyen Thi Thu Ha, 2001). All these human activities have resulted in changes of land and forest cover in the area. Forest cover change can happen in many ways. It can be a degradation process if forest quality or forest ecological function declines. It can also be either a re-growth or deforestation process. The FAO (2000) has defined deforestation as the permanent change of land use from forest to other type(s) of land use or the depletion of forest crown cover to less than 10 percent. However, the meaning of deforestation adapted to land cover and/or land use mapping is very different in various countries. In Vietnam, according to FIPI, deforestation simply means the disappearance of dense forest trees, which consequently leads to the decrease of tree cover and the depletion of forest ecological functions. The need for land use and land cover change information has grown steadily since the late 1990’s when priority was shifted to setting up long-term management strategies for natural resources. Many studies such as those by Chen (2000), Diouf & Lambin (2001), Kuntz & Siegert (1999) have emphasized the importance of investigating land cover dynamics as a baseline requirement for sustainable management of natural resources. The ability to answer the questions “where are the changes” and “what are causes of the changes” is essential for the formulation of appropriate management strategies. The understanding of land cover change and/or the forest cover change process and its underlying causes will help government policy makers and resource managers to decide on where action should be taken and what kind of intervention is needed. However, despite ongoing efforts, there is little information about land cover dynamics, especially with regards to forest cover, and their driving forces in the Ca River Basin. This study’s aim, therefore, is to investigate the implications of the region’s biophysical conditions, its socio-economic context, and the implementation process of the government’s policy on land allocation. More specifically, the objectives of the study are (i) to estimate the rates of forest cover changes in the upper Ca River Basin during the period 1998 - 2003 and (ii) to determine the main socio-economic and biophysical factors governing forest cover changes in the period 1998 - 2003. 2. MATERIALS AND METHODS Study Site The main study site is located in the upper part of the Ca River Basin, which covers a vast area of the Tuong Duong and Ky Son districts. Due to the availability of satellite images and statistical data, 41 communes were analyzed. A map view of these communes is shown in Figure 1. 32 Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam N ghe An Upper Ca River Basin Ky Son and Tuong Duong Figure 1. Study Area, located in the Upper Ca River Basin. Land Cover / Land Use Mapping Land cover mapping has become one of the most important and typical applications of remote sensing. It is an integrated process, often known as a classification system, based on the identification of levels and classes. The level and class should be designed in consideration of the purpose of use (national, regional or local), the spatial and spectral resolution of the remote sensed data, user’s request and so on (Japan Association of Remote Sensing, 1996). According to Jensen (1996) there is a fundamental difference between information classes and spectral classes. Information classes are those defined by man while spectral classes are those inherent in the remote sensing data and must be identified and labeled by the analyst. The aim of digital classification is to translate spectral classes into information classes. Two sets of ETM images were used to map the land cover of the period 1998 - 2003. The images captured the study site in the dry season, once in May 1998 and the other in April 2003. All images were co-registered into each other and in WGS 84 Datum and zone 48N. Prior to the classification process, a low pass convolution filter with a filter window of 3x3 was applied to all images, as suggested by Tottrup (2001). This helped to smooth images and diminished the terrain effect on the surface reflectance in order to gain a better land cover mapping. Moreover, experience gained by working with satellite images gathered during the region’s dry season has shown that with quite limited ground truth points, it is very difficult for interpreters to distinguish spectra differences among several objects, such as dry paddy fields, build-up areas and swidden fields. Therefore, though the training samples were taken toward very diverse land cover types, the final land cover categories have been grouped in five major classes as shown in Table 1. This also allowed for improving the accuracy assessment of the land cover/land use map later. 33 Nguyen Thi Thu Ha Table 1. Land cover/land use mapping category. LC category Primary forest Degraded forest Karst ( * ) Bamboo Fallow Agriculture Water Cloud 1998 x x - x x x x x 2003 x x x x x x x x Description Less accessible by humans with very dense and tall trees Logged, regenerated and secondary forest Mature, young and planted bamboo Bush, grass mixed with small trees Paddy, swidden and bare ground Rivers, lakes, ponds, etc. Masked Note that the Karst could not be mapped well in 1998 due to the mix of its spectra library with that of the degraded forest. However, this would not affect the later forest cover change analysis as Karst was excluded from the target land cover groups. The land cover mapping was performed in the ENVI 4.2 environment with the maximum likelihood function. Accuracy Assessment for Land Cover Mapping In order to assess the accuracy of the 1998 map, two sets of ground truth points collected by Tottrup in 2000 and Leisz in 1999 were used. For the 2003 analysis, one set of ground truth points collected surrounding the area of Luu Kien commune was used. Points already used to train the sample sets for maximum likelihood classification were excluded in this procedure. The most common use for accuracy assessment is Kappa statistics which is calculated by using Equation 1 (Jensen, 1996) ∑ ∑∑ = ++ == ++ − − = r 1i ii 2 r 1i r 1i iiii XXN XXXN k ˆ where: “r” is the number of rows in the error matrix, X ii is the number of observations in row i and column i, and X i+ and X +i are the marginal totals for row i and column i, respectively, and N is the total number of observations. Kappa statistics were also used in assessing how well the training sets match the classification. The assessment was carried out using function Confusion matrix using ROI ground truth in ENVI. Change Detection with Post-Classification This technique in ENVI allowed generating a matrix table, which reflects the land cover change between 1998 and 2003, and “change” maps corresponding to selected land cover categories. The matrix table was then used to calculate the rate of change under the forest cover type for the period. However, since the analysis later focused on the forest cover dynamic and its underlying causes, one intermediate step had been taken to reclassify the change detection maps into a new map that was set up with three major forest change types. The rules are in Table 2. (Equation 1) 34 Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam Table 2. Land cover change detected by the post-classification method. No. Land Cover 1998 Land Cover 2003 Regrouping Primary forest Degraded forest 1 Degraded forest Bamboo Fallow Agriculture Deforestation Degraded forest Primary forest 2 Bamboo Fallow Degraded forest Forest regrowth Primary forest Primary forest 3 Degraded forest Degraded forest No change Bamboo Fallow Agriculture Fallow Bamboo Agriculture Agriculture Fallow Bamboo 4 Cloud, water, Karst Other land use types Not considered or unidentified Logistic Regression in SPSS Software The logistic binary regression technique in the SPSS statistical package version 15.0 was used to investigate the relationship between biophysical and socio-economic factors and forest cover changes. The nature of forest cover change variables was considered to be binary i.e. change or no change. They formed the dependent variables in the analysis while biophysical and socio-economic factors served as independent or explanatory variables. The analysis was carried out to investigate if the association between the underlying factors and land cover changes were consistent over time. The analysis was followed by stepwise- forward conditional interactions in SPSS 15. Dependent variables, here the forest cover changes in Table 1, were then recoded into 0 and 1 with representative of no change and change (forest regrowth and deforestation). Several independent factors were selected for the regression analysis as shown in Table 3. Table 3. Independent Factors for Logistic Regression Analysis. Independent factors Unit Source Slope Degree Contour map/DEM Elevation 100m Contour map/DEM Implementation process of the land allocation policy 0-1 Secondary data plus official interviews Population density Number of people per sq. km Statistical data Cattle density Number of cattle per sq. km Statistical data Food security Crop production per person Statistical data Distance from roads 500m Buffer operation in GIS Distance from river 500m Buffer operation in GIS In order to use effectively the binary logistic regression, three thousand random points were taken within the boundary of the study area. These points were then rasterized and overlaid on each individual determinant factor map together with the final forest cover change map. The ILWIS 3.3 cross function was performed to retrieve all information at each randomly selected point. In the end, 183 points that satisfied the requirement were taken into the logistic regression. 35 Nguyen Thi Thu Ha Processing GIS operation ETM 1998 ETM 2003 data Social data Land cover change map 1998-2003 Change detection Height, slope and distance maps Generating dependent variables Generating explanatory variables Dependent variables Explanatory variables Multiple logistic regression Explaining models of forest cover change Random sample points Biophysical Figure 2. Schematic Diagram of the Research Method. 3. RESULTS Forest Cover Change Land cover/Land use mapping Figure 3. Land cover/land use 2003 36 Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam The results of land cover mapping are shown in the following Figure 4. As the study’s focus is on forest resources, only five land cover types will be analyzed. The others will not be taken into account as they are not involved in the logistic regression analysis. Table 4 illustrates the area and percentage of the five different land cover types. Area of land cover types 1998 2,040 59,645 171,593 104,069 117,789 94,362 24,837 0 25,000 50,000 75,000 100,000 125,000 150,000 175,000 200,000 225,000 250,000 Water Agr i c u ltural land Fallow Bambo o Degraded forest Pr i mary fo r es t Clou d Area_ha Area of land cover types 2003 4,785 68,866 213,352 52,239 114,316 115,792 23,950 4,632 0 25,000 50,000 75,000 100,000 125,000 150,000 175,000 200,000 225,000 250,000 W at er Agric ul tural la nd Fall o w Bambo o Degraded fores t Pr i mary fore s t C l oud Karst Area_ha Figure 4. Maps and areas of different land cover maps for 1998 and 2003. Table 4 Area of Land Cover Types (ha). Land cover 1998 % 2003 % Fallow 170,128 27.9 198,998 32.7 Bamboo 102,136 16.8 51,021 8.4 Degraded forest 110,450 18,1 109,345 18.0 Primary forest 84,218 13.8 102,678 16.9 Agricultural land 57,809 9.5 62,698 10.3 37 Nguyen Thi Thu Ha Table 4 provides the general trend of the 5 major land cover types over the period. The fallow area actually increased, showing that over 5 years the area opened for agricultural land had increased. That trend matches with the difference of agricultural land area in 1998 and 2003. Area under primary forest cover increased, reaching about 3.1% in 2003, while the percentage of degraded forest was fairy stable. The reason behind this is that some degraded forest area has been converted to agricultural area, but the bamboo and fallow might turn into degraded forest. This is an example why change detection is very helpful. Accuracy assessment for land cover mapping Accuracy assessment for land cover maps was performed by using the confusion matrix. Apart from this, Jeffries-Matusita’s separability was carried out to assess the training samples for the maximum likelihood classification. Table 5 is the Jeffries-Matusita’s separability for the training samples of 1998 and 2003. The Jeffries-Matusita’s value ranges from 0 to 2, and if the Jeffries-Matusita’s value of one class pair ≥ 1.9, the classes have very good separability. Table 5. Accuracy Indices for Land Cover Maps of 1998 and 2003. 1998 Overall Accuracy = (178/226) 78.8% Kappa Coefficient = 0.72 Class Agriculture Fallow Bamboo Degraded forest Primary forest Prod. Acc (%) 88.14 86.11 60.71 70.37 91.67 User Acc. (%) 83.87 72.09 79.07 86.36 84.62 2003 Overall Accuracy = (146/181) 80.7% Kappa Coefficient = 0.72 Class Agriculture Fallow Bamboo Degraded forest Primary forest Prod. Acc (%) 88.24 75.00 89.47 52.00 100 User Acc. (%) 83.33 76.74 65.38 92.86 100 Detected changes Change detection maps provided in ENVI are very detailed at eight land cover types (according to the land cover map of 2003). However, as explained in the method, the final produced map for forest cover change will consist of only three major categories: forest re- growth, deforestation and no change. The result is shown in Figure 5a & b. 38 Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam 47,730 76,467 52,688 66,833 7.3 11.7 8.1 10.2 0 10,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 90,000 Deforestation Forest regrowth No change for degraded forest No change for primary forest Area_ha 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 Rate of change (%) Figure 5a. Area of Forest Changes (ha) and Rate of Change to Total Area. It can be seen in Figure 5b that forest re- growth mostly occurred within the boundary of Pu Mat National Park, along road No.7 and along the part of the Ca River belonging to Tuong Duong and Con Cuong districts. In the northeastern part of the region, toward the boundary of Pu Huong National park, deforestation appears more frequently. Two other places where more deforestation happened are Tam Hop, Tuong Duong and Na Ngoi, Ky Son. Relationships between Change and Determinant Factors Recoding of dependent variables and categorical explanatory variables was necessary for the logistic regression analysis. Two major types of changes are taken into analysis, deforestation and forest regrowth. They are recoded into binary variables 1 and 0 representing “change” and “no change” respectively. The categorical explanatory variable management effect denoted as MANAGEMENT is as recoded 1 and 0, representing area where land allocation policy was already implemented, and for area where the policy hasn’t been yet processed, respectively. Figure 5b. Change Map by Post Classification, 1988-2003. 39 Nguyen Thi Thu Ha Table 6. Recoding variables. Variables Dependent Recoding Change No change Forest regrowth 1 0 1 0 Deforestation Independent With land allocation policy No land allocation policy Management 1 0 Forest Regrowth Analysis Also, prior to the logistic regression exercise, collinearity tests were performed for all independent variables (see Table 3). The tests showed no collinearity with the tolerance ranging from 0.42 to 0.69, which is higher than the critical value of 0.2. Therefore, all the independent variables were used in the multiple logistic regression analysis. Table 7. Factors Significantly Associated with Forest Regrowth. Variables Unit B S.E. Wald df p_value Exp(B) Pop_den Number of people/km 2 .325 .143 5.183 1 .023 1.385 Cow_den Number of livestock /km 2 -1.258 .494 6.475 1 .011 .284 DEM 100m .008 .003 5.957 1 .015 1.008 Constant 16.203 5.528 8.591 1 .003 1E + 007 Form Table 7 it can be seen that there are three factors associating with forest regrowth. The elevation (DEM) and population density are positively related to natural forest regrowth. This means that the odds for forest regrowth will increase 33% when the population density increases; and the odds for forest regrowth will increase by 1% with every unit of 100m elevation increase. The most predictive factor for forest regrowth is livestock density, with the Wald value of 6.5. With the negative intercept at 1.253, it can be interpreted that the odds for forest regrowth will increase 1.2 times if the cow density decreases. The model for forest regrowth derived from table 9 is )X08.0X258.1X325.0203.16exp(1 )X08.0X258.1X325.0203.16exp( P 321 321 REGROWTH +−++ +−+ = where: P REGROWTH is the probability of forest regrowth X 1 is the population density (people/km 2 ) X 2 is the livestock density (number of cows/km 2 ) X 3 is the elevation (100m) The goodness of fit for the model is . This is model with very good fit. 68.0 2 = L R Deforestation analysis Table 8 below provides another look at forest cover change in the Ca River Basin. Deforestation during the period 1998-2003 shows that three factors (food security, management and livestock density), are all negatively related to deforestation. However, the livestock density factor is the least effective factor with the Wald value of 8.5 and the intercept B of 0.154. 40 [...]... Allocation Policy, food security, and livestock density in the deforestation model These predictors have built up a very good logistic model for forest cover changes with 2 the RL ranging from 0.22 to 0.68 Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam REFERENCES Chen, X (2000) Using remote sensing and GIS to analyse land cover change and its impacts on the regional sustainable... additional activities in order to sustain their lives Hence, areas where food security is much lower than livestock number could be one cause of deforestation, whilst the other could be illegal logging However, we should keep in mind that in general in the Ca River Basin, not only population density, but also livestock density hasn’t exceeded the region’s natural carrying capacity, and therefore they may not... regression analysis for forest cover change in the Upper Ca River Basin has shown that, conversely with what people often think, forest cover change doesn’t occur often near roads or rivers This case is somewhat contrary to a PREGROWTH = case study in Kenya where road accessibility played a contributing role to deforestation (Serneels & Lambin, 2001); even with a case study in Bach Ma National Park... areas in Ky Son and who are more dependant on forest resources In both models for forest regrowth and deforestation, the number of livestock per area was found in association with forest change While in the case of forest regrowth, the livestock effect is very clear, in the case of deforestation it is not so readily discernable This is perhaps due to food security, another factor associated with deforestation... represent “causes” to either deforestation or forest regrowth Further studies with more determinants, such as distance from village, type of agriculture practice, or household economy could help us to better understand the underlying causes behind forest cover change in the region 5 CONCLUSIONS Over the study period from 1998 to 2003, the change rates of forest cover were found to be 11.7 and 7.3% for forest. .. and deforestation, respectively The analysis for the driving forces to these changes by using the multiple logistic regression technique showed that the Land Allocation Policy and natural management practices were the most important factors These are reflected through the number of livestock per area, population density, elevation in the forest regrowth model, and the implementation process of the Land.. .Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam Table 8 Factors Significantly Associated With Deforestation Variables B S.E Wald df Sig Exp(B) Food_sec -.016 Management(1) -1.577 005 8.679 1 003 984 478 10.874 1 001 207 Cow_den -.154 053 8.475 1 004 857 Constant 4.799 1.151 17.378 1 000 121.433 The interpretation for the food security Reading the most effective... PDEFOREST is the probability of deforestation X1 is the food security (total crop production in kg/person) X2 is the management (for Land Allocation Policy) X3 is the livestock density (number of cows/km2) 41 Nguyen Thi Thu Ha There is a small surprise found in the model of forest regrowth here This is the positive relation of population density with forest regrowth This is probably because: (i) The. .. PDEFOREST = 1 + exp(4.799 − 0.016X 1 − 1.577 X 2 − 0.154X 3 ) where: PDEFOREST is the probability of deforestation X1 is the food security (total crop production in kg/person) X2 is the management (for Land Allocation Policy) X3 is the livestock density (number of cows/km2) 2 The goodness of fit for the deforestation is RL = 0.22 , which is a model with moderate fit 4 DISCUSSION The results of the. .. the most effective factor to is that the odds for deforestation will deforestation, the management factor, it is very clear that the change to deforestation in the area where increase about 2% if the food security land allocation policy has not been implemented is decreases 21% higher than the deforestation in the area where The model for deforestation is land allocation policy was already launched exp(4.799 . 32 Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam N ghe An Upper Ca River Basin Ky Son and Tuong Duong Figure 1. Study Area, located in the Upper Ca River. Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam The results of land cover mapping are shown in the following Figure 4. As the study’s focus is on forest resources,. 1) 34 Driving Forces of Forest Cover Dynamics in the Ca River Basin in Vietnam Table 2. Land cover change detected by the post-classification method. No. Land Cover 1998 Land Cover 2003